chore(telemetry): clean up what release 1 left on next - #1053

🎯 What & why

Cleans up what the telemetry release 1 work left on next: code for a ledger format no release wrote, probes and history in task folders, docs that drifted from the code, and two weak guard tests. Each audit item is one commit. Net: 44 files, +153 / -2043, 17 files deleted.

🛠️ How it works

Monthly ledger: why it is dead.

• Only next builds between feat(telemetry): deterministic claude code token measurement, release 1 #993 and perf(telemetry): the ledger is kept by day, so an ingest costs what it adds #1031 wrote a month file.
• cli-v5.5.0 is the first release with the ledger, and its adapter already writes one file per day.
• cli-v5.4.0 had no ledger in ledger/; its day files went to another sink.
• fix(telemetry): one call, one key; one repository, one id #1049's migration never reads or writes a month file.
• Effect on a stray 2026-10.jsonl in ledger/ now: it is not a partition, so it is neither read nor removed. forget still removes the whole directory.

Tests changed, not only deleted (item 2).

• ledger-upkeep.property, report-figures.property and report-window.property kept their starting-store dimension. It is now "days an earlier ingest wrote, unfolded" instead of "a month file", so the coverage of a pre-existing store stays.
• stored-usage-migration.property lost its October-is-monthly flag. Its docstring already says "as cli-v5.5.0 left them".
• The unsettled-partition ingest test and the forget test now use a day name.

Guard change inside item 5.

• scripts/__tests__/architecture-doc-matches-the-tree.test.js allowed exactly one script per plugin row, which is how the table came to name only prompt-gate.cjs.
• It now checks that the Runs cell equals every script the plugin's hooks.json runs, and that each script exists.
• Red against the old table: actual: [ 'hooks/prompt-gate.cjs' ], expected: [ 'hooks/catch-up.cjs', 'hooks/prompt-gate.cjs', 'hooks/session-start.cjs' ].

🧪 How to verify

• pnpm exec lefthook run pre-commit --all-files exits 0.

• pnpm exec lefthook run pre-push --all-files:

  • knip is green;
  • cli suite: Test Files 522 passed | 1 skipped, Tests 6787 passed | 1 skipped.

• Bundle (pnpm build): 700.6 KB on origin/next, 699.8 KB on this branch.

• Test-first:

  • The adapter test reads only day files, and leaves a month's name, ... alone failed before the change with skippedLines: 1, damaged: {"2026-10"}, and passes after it.
  • Hand mutation: deleting the window test in load turns 4 adapter tests red.

• Same store, same report. Steps:

  1. In a sandbox (HOME, XDG_CONFIG_HOME, AIDD_USER_CONFIG_DIR, AIDD_TELEMETRY_DIR, CLAUDE_CONFIG_DIR in temp dirs, GIT_CONFIG_NOSYSTEM=1), npm i @ai-driven-dev/[email protected].
  2. aidd telemetry on, then backdate the consent, the same way the e2e tests do.
  3. Ingest the repository's cli/tests/fixtures/claude-usage transcripts with 5.5.0.
  4. Copy the store, then run report --json with the origin/next build on one copy and this branch's build on the other.

  Result, for no period, --from 2026-10-07 --to 2026-10-07 and --axis day: the reports are byte-identical, and so are the stores they leave behind.

  Limit: the fixture store is one day and 4 records.

• Guard nits, by hand mutation:

  • ANY_UUID without i turns reads every session array ..., whatever the case of its ids red.
  • continue instead of return on an unclosed array turns stops at the first unclosed session key ... red.

⚠️ Heads-up

• Not done:
  • Audit item 1 (the leak) shipped in chore: remove local machine paths from the repo and guard against new ones #1051.
  • Audit item 11 (trimming evidence and reviews) was excluded: those files are kept.
  • In item 6, the bounded-read bullet stays in the contract rather than pointing to the Stores row. It states which reads print the same figures; the Stores row states the file operations.
• Dropped from the audit:
  • "done and implemented make two columns" is wrong: kanban/src/domain/models/progress-status.ts maps implemented and reviewed to Done.
• Kanban gaps, not fixed here:
  • framed, approved and superseded are unmapped and land in the unknown column.
  • Both status edits are sub-documents under a plan.md already in Done, so the board does not move.
  • deterministic-tokens/context/ still holds three docs with no status, so it remains a stray unknown card.
• Text mentions left as history (not links):
  • transcripts-are-read-as-a-stream/review.md still cites memory-bench.mjs lines.
  • The Decisions of deterministic-tokens/plan.md cite "phase 6 task 5" and "phase 4 task 3.2".
• Mutation: Stryker was not run locally; CI ran the telemetry mutation shards. Item 2 deletes a dead branch and inlines the window test; the hand mutation above covers the line that changed.
• aidd-dev:01-plan still generates phase-N.md files, which the new task-folder policy folds into plan.md. Changing that skill is a product decision left to the maintainer; plugins/aidd-dev is untouched.
• Audit item 9 (fallback key), out of scope: fix(telemetry): one call, one key; one repository, one id #1049 dropped the timestamp from the requestless key. A copy written under another session id would still not fold, and the Residual limits table does not say so.

🔗 Linked issue

None. Found by the R1 telemetry cleanup audit.

✅ I certify

• I DO CERTIFY I READ EACH LINE OF THE PULL REQUEST BECAUSE I AM A SOFTWARE ENGINEER, NOT A AI PUPPY.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VWNxk63AGKkqE8HRqHLjGp